Energy Efficient Ultra-Dense Network Using Long Short-Term Memory

Abstract

The energy consumption of cellular systems is becoming a matter of grave concern in both economic and environmental perspectives. Recently, in order to reduce the energy consumption of base stations (BSs), which takes the largest portion, turning off under-loaded BSs has been suggested. However, determining the on/off mode of BSs is a non-convex optimization problem. Also, the problem must be solved in accordance with the time-varying environment since the transition overhead in the future may outrun the power saving at the moment. In this paper, we propose Long Short-Term Memory (LSTM) based framework to make far-sighted control decisions maximizing energy efficiency from a long-term perspective. The LSTM-based network can intelligently determine the on/off modes, utilizing the time-correlated property of the channel and approximating complex mapping between channel state and desired power control coefficient. Lastly, through the convex optimization technique, the optimal power allocation for the active BSs can be found. Simulation results show that the proposed technique outperforms the conventional techniques by a large margin.